Spatio-temporal feature fusion for real-time prediction of TBM operating parameters: A deep learning approach

This research provides a spatio-temporal approach to perform real-time forecasting for the tunnel boring machine (TBM) operating parameters. By extracting the real-time TBM operational data from the data acquisition system, a Long Short-Term Memory (LSTM) based deep learning model is trained for accurate prediction. A global sensitivity analysis (GSA) by adopting the Sobol method is performed for the model to quantify the contribution of input variables. The developed methodology can be a useful tool for TBM performance improvement and it enhances the state of knowledge on underground excavation. The result from the case study indicates that: (1) The proposed spatio-temporal method provides reliable real-time forecasting with mean absolute error (MAE) and root mean squared error (RMSE) of 1.261 mm and 1.955 mm, respectively, and (2) GSA results indicate that TBM's thrust and CHD torque are the 2 most influential spatial factors, while the historical data of penetration rate is critical for accurate forecasting. Further studies could focus on backward optimization to improve TBM's performance based on the prediction.

Automated summary

This research presents a spatio-temporal approach for real-time forecasting of TBM operating parameters, using a deep learning model trained on real-time operational data. Global sensitivity analysis using the Sobol method identifies thrust and torque as the most influential factors, with historical penetration rate data critical for accurate forecasting. Further studies could focus on backward optimization to enhance TBM performance based on prediction results.